Condition-dependence of multiple carotenoid-based plumage traits: an experimental study


  • Anne Peters,

    Corresponding author
    1. Behavioural Ecology of Sexual Signals Group, Vogelwarte Radolfzell, Max Planck Institute for Ornithology, Schlossallee 2, Radolfzell 78315, Germany; and
      *Correspondence author. E-mail:
    Search for more papers by this author
  • Kaspar Delhey,

    1. Behavioural Ecology of Sexual Signals Group, Vogelwarte Radolfzell, Max Planck Institute for Ornithology, Schlossallee 2, Radolfzell 78315, Germany; and
    Search for more papers by this author
  • Staffan Andersson,

    1. Animal Ecology, Department of Zoology, Göteborg University, Box 463, 405 30 Göteborg, Sweden
    Search for more papers by this author
  • Hendrika Van Noordwijk,

    1. Behavioural Ecology of Sexual Signals Group, Vogelwarte Radolfzell, Max Planck Institute for Ornithology, Schlossallee 2, Radolfzell 78315, Germany; and
    Search for more papers by this author
  • Marc I. Förschler

    1. Behavioural Ecology of Sexual Signals Group, Vogelwarte Radolfzell, Max Planck Institute for Ornithology, Schlossallee 2, Radolfzell 78315, Germany; and
    Search for more papers by this author
    • Present address: Institute of Avian Research, An der Vogelwarte 21, Wilhelmshaven, Germany.

*Correspondence author. E-mail:


  • 1Condition-dependent expression of ornamental traits is a fundamental assumption of theories on the honesty of sexual signals, and it is widely assumed that condition-dependence is a signature feature of ornaments.
  • 2Some of the best understood condition-dependent traits are the striking carotenoid-based plumage signals of male birds, yet little is known about the many less conspicuous, less elaborate carotenoid-based plumage colours that often comprise large parts of the plumage.
  • 3We examined colour (reflectance) of carotenoid-based plumage in male greenfinches that were provided with naturalistic diets with relatively low and with enhanced lutein availability during their annual moult. Using a variety of objective colorimetrics, including physiological models of avian colour vision, we compared experimental effects and general condition-dependence on the contrasting bright yellow tail patch, the yellow-green breast as well as three duller, yellow- to olive-green patches (back, crown, rump).
  • 4Irrespective of the analysis method used, we found consistent and large diet effects on the reflectance of the tail, much weaker effects on the reflectance of the breast, and no significant effects on the other three plumage parts. Likewise, we found that only the colour of the tail was strongly associated with circulating (plasma) lutein concentration, as well as with general condition (body mass, haematocrit).
  • 5Our results suggest that, in accord with current theories on the signal honesty, the striking yellow tail patch of the male greenfinch appears to be particularly well-adapted to signal information on carotenoid availability and general condition of the male during moult.


Handicap models of evolutionary maintenance of exaggerated sexual ornaments state that such elaborate traits are costly and thereby reliable indicators of general health or quality (Andersson 1994). Condition-dependent expression of such traits is a fundamental assumption of theories on the honesty of sexual signals, and it is widely assumed that condition-dependence is a signature feature of ornaments (Johnstone 1995). However, to what extent condition-dependence is particular to elaborate ornaments compared to similar less elaborate, non-signalling traits remains contentious (Cotton, Fowler & Pomiankowski 2004b).

Carotenoid-based ornaments are among the most straightforward candidate examples of condition-dependent elaborate traits (Hill 2006). Animals can only obtain carotenoids through their diet, and the amount deposited in ornaments may thereby reflect the individuals ability to acquire and assimilate these pigments (McGraw 2006). Moreover, carotenoids are also antioxidants and immunomodulators and more colourful individuals may directly as well as indirectly signal their health (reviews in McGraw 2006; Peters 2007). For example, carotenoid-based colour of feathers is not only affected by carotenoid content of the diet and to a lesser extent by general condition (Shawkey et al. 2006) but also by parasite burdens (Hõrak et al. 2004). For this reason, carotenoid-based ornaments have featured prominently in research on honesty of sexual signals (Cotton et al. 2004b; Hill 2006).

Carotenoids, particularly the yellow xanthophylls, are not only deposited in contrasting and bright plumage patches, but also in less conspicuous plumages, often together with melanins, to produce duller, more cryptic yellow-green, green and grey–green plumage (Lucas & Stettenheim 1972). Theoretically, one would expect the eye-catching signalling plumage to be a better indicator of condition and carotenoid availability (Cotton et al. 2004b) but no study has compared carotenoid- or condition-dependence of elaborate, bright and more cryptic, dull carotenoid-based traits.

We examined the effect of moderate carotenoid supplementation on carotenoid-based plumage expression in male European greenfinches Carduelis chloris. Greenfinches display a variety of carotenoid-based plumage patches, ranging from bright yellow to olive. The contrasting yellow tail patch, and to a lesser extent the yellow-green breast, are known indicators of immunocompetence and parasite resistance (Lindström & Lundström 2000; Saks, Ots & Hõrak 2003b; Hõrak et al. 2004). We examined the effect of carotenoid supplementation during moult on these two relatively bright plumage patches as well as three duller patches (back, crown, rump) by analysing reflectance spectra using physiological models of avian colour vision. Additionally, we determined the relationship between mid-moult circulating levels of carotenoids and general condition on expression of plumage colour.


study species

Greenfinches are sexually dichromatic cardueline finches with males being yellow-green and olive and females yellow-buff and olive-brown (Cramp & Perrins 1994). They undergo one annual moult during late summer and autumn when they replace their entire plumage. Both sexes show contrasting yellow patches on the primaries and the side of the tail due to deposited hydroxy-xanthophylls (lutein and canary xanthophylls A and B, derived from lutein through dehydrogenation, McGraw 2006). The yellow-green to olive colour of the rest of the plumage is presumably the result of co-deposition of melanin (primarily in barbules) together with the yellow xanthophylls (Lucas & Stettenheim 1972). Reflectance analysis confirmed discriminable sexual dichromatism of all carotenoid-based plumage (using methods similar to those described below) during the breeding season, with the breast being most dichromatic, followed by the tail, head, back and finally the rump (Delhey & Peters 2008).

The expression of carotenoid-based plumage coloration in male greenfinches is under sexual selection (Eley 1991). Eley (1991) visually assessed ‘brightness’ of different plumage patches – including the intensity of yellow in the tail, belly and wings, the length of the yellow parts of the tail and wing feathers, the greenness of crown and rump plus several other variables. From this set of concordant variables she computed a summed Z-score as a composite measure of brightness. Males with higher scores, that is, those that were overall ‘brighter’, had greater reproductive success as they had more (social) mates and reared more chicks per brood and recruited more offspring to independence. However, brighter males were at increased risk of Sparrowhawk predation and injuries, indicating that coloration could be a sexually selected handicap (Eley 1991).

A number of studies have investigated potential signalling and indicator roles of greenfinch plumage. Merilä, Sheldon & Lindstrom (1999), using a similar composite index of overall plumage ‘yellowness’ based on percentage of yellow in the wing and intensity of yellow in wing, tail, belly and breast, were the first to demonstrate that male plumage yellowness varies with blood parasite infection and can thus function as an indicator of male quality. Further studies examined different plumage regions in more detail, particularly the breast and tail plumage.

The bright, contrasting yellow tail patch appears to be the most established indicator of health and immunocompetence. It is the most sexually dimorphic, being larger in males to the extent that it can be used to assign sex with > 95% accuracy (Cramp & Perrins 1994). Males with more yellow in the tail (estimated as the main factor contributing to a principal component of wing, tail, and breast plumage conspicuousness) have higher virus clearing capacity (Lindström & Lundström 2000). Moreover, chroma and hue in the visible range of the spectrum (see below) of the colour of the tail were depressed by coccidian infection during moult (Hõrak et al. 2004).

The breast is greenish-yellow, paler and less uniform in females (Cramp & Perrins 1994). Reflectance of breast feathers in males showed some sensitivity to coccidian infection, but less than the tail (Hõrak et al. 2004). In one study, breast chroma (saturation between 400 and 700 nm) was negatively related to humoral immunocompetence (Aguilera & Amat 2007), while in another, breast brightness (total reflectance between 400 and 700 nm) was not related to cellular immunity and positively related to humoral immunocompetence (Saks et al. 2003b). However, as carotenoid (as other pigment)-based colours are subtractive (i.e. more deposited pigment absorbs more light, Andersson & Prager 2006), this also implies that more immunocompetent birds had less carotenoids in the feathers. Therefore, at this stage we conclude that there is mixed evidence that breast colour can signal male quality.

Crown, back and rump are greenish-yellow to olive-green (Cramp & Perrins, 1994), and no specific information on potential signalling function is available, in accord with the general paucity of studies focussing on evolutionary function of less colourful plumage. Greenfinches also show a bright yellow patch in the wing, that is an indicator of age (Cramp & Perrins 1994; Lindström & Lundström 2000). We used only adult male birds and they showed little variation in yellow wing patch (measured according to Lindström & Lundström 2000, data not shown).

experimental design

Between 8–13 July 2005 we captured 20 adult male greenfinches near the Vogelwarte Radolfzell, Germany (at this time of year it is not possible to unambiguously distinguish between last year and older birds, Cramp & Perrins 1994). Birds were ringed, and we measured plumage reflectance (see below), body mass and tarsus. After a brief (1–2 days) habituation period in small cages, birds were housed in large individual outdoor aviaries of 3·0 × 3·0 m wide and 1·9 m high, planted with native shrubs. Greenfinches in the wild consume a great variety of seeds (Cramp & Perrins 1994). We provided birds daily with ad libitum fresh water plus a standard diet consisting of 15 g unpeeled sunflower seeds (carotenoid content 1 µg g−1, (McGraw et al. 2001)), 3 g canary seed mixture (Witte Molen, Netherlands, carotenoid content 2·8 µg g−1, (Hudon 1994)), 2 g of a nutrient rich food mix (containing crushed rusk, wheat germ, milk-powder, soy flour, oat, fructose and egg-whites (8·1% carbohydrates, 4·0% proteins and 2·7% lipids per wet mass) with complete vitamins Vitakalk and Korvimin (neither contained carotenoids) added according to manufacturers instructions) mixed together with one teaspoon sesame oil. Although we could not determine exact food consumption, birds almost exclusively consumed the sunflower seeds, their preferred food at feeders and in captivity (Cramp & Perrins 1994) and their staple or sole food in other captive studies (Saks, McGraw & Hõrak 2003a; Hõrak et al. 2004; Karu, Saks & Hõrak 2007).

After 1–2 weeks in captivity (20 July) we confirmed that birds had not yet started to moult and took a small (pre-moult) blood sample for haematocrit and plasma analysis. Then (21 July), we started the experimental diet treatment, randomly assigning half of the birds to receive an added supply of carotenoids (0·2 g per day per bird of Lutein 5% beadlets, Pfannenschmidt Hamburg) mixed into the standard food (experimental group). Progress of moult (all birds were undergoing a complete moult) and bird condition was inspected around mid-moult (23 September) and a small blood sample was taken. Additionally, on 8 and 18 August we captured and took a blood sample of six free-living moulting adult male greenfinches. When moult was complete, 2–3 November 2005, all birds were again colour measured, a small post-moult blood sample was taken, and all birds were released in good health in the area of capture. The experiment was performed under licence from the Regierungspräsidium Freiburg (Nr. 55-8852·15/05).

reflectance measurement and analysis

For each bird we obtained measurements of plumage coloration before the start of moult and after moult was completed. We measured reflectance between 300 and 700 nm of the following regions: crown, back, breast, rump and yellow part of the tail feathers. We used an Avaspec 2048spectrometer, a DH-S light source and a bifurcated fibre optic probe (all Avantes, The Netherlands). The probe, with a black plastic cylinder (‘probe holder’, Andersson & Prager 2006), at the tip to standardize measuring distance and exclude ambient light, was placed perpendicular to the surface, hence illumination and recording angles were both 90° (the ‘Coincident Normal’ configuration in Andersson & Prager (2006)). Reflectance was computed relative to a WS-2 white standard using the program Avasoft 6·2·1. We took a set of five reflectance readings of different predefined and standardized spots (each around 11 mm2) in each body region. These reflectance spectra were then imported into Excel for further analysis.

For each reflectance plumage patch we determined cone quantum catches. Based on spectral sensitivities of the four cones (VS: very short, S: short, M: medium, and L: long wavelength sensitive cones) used in colour vision (Vorobyev et al. 1998b), we calculated standardized cone quantum catches (Qi) using the formula: Qi = ∫λRi(λ) S(λ) I(λ) dλ, where λ indicates wavelength, Ri(λ) the sensitivity of the cone type (cone sensitivities for type U-eyes according to Appendix A in Endler & Mielke ( 2005)), S(λ) the reflectance spectrum, and I(λ) the spectrum of irradiant light (D65 standard daylight, Vorobyev et al. 1998b). We then divided each cone quantum catch by the sum of all four and transformed these according to Kelber et al. (Kelber, Vorobyev & Osorio 2003), to obtain three independent relative cone catches, x, y and z, whereby, according to this transformation, higher values of x represent greater stimulation of the L cone and lower stimulation of the M cone, higher y values represent greater stimulation of the S cone, and higher values of z greater stimulation of the VS cone. These relative cone catches can be represented as points in the tetrahedral avian visual space and we calculated chroma as the Euclidean distance of each point to the theoretically presumed achromatic centre (where each cone is equally stimulated, Kelber et al. 2003). Relative cone catches were moderately to highly correlated: correlation coefficients | r | (calculated separately for pre-moult, post-moult and the change over moult and per patch) between x, y and z averaged 0·66 ± 0·04 (mean ± SE, range 0·02–0·97). For each set of x, y, and z we calculated the first principal component (PCxyz), that explained 47–96% (mean ± SE = 77·1 ± 3·8) of the variation. PCxyz was correlated with Euclidean chroma (mean | r | ± SE = 84 ± 6).

Additionally we calculated carotenoid chroma (R700 – R450)/R700 from the reflectance spectra. This measurement most accurately reflects carotenoid content of feathers in unsaturated colours (i.e. ‘for which not all reflectance has been reduced in the spectral region of maximum absorptance’Andersson & Prager 2006). Carotenoid chroma was correlated with PCxyz (mean | r | ± SE = 77 ± 7).

Finally, we assessed whether average colour differences between diet groups could be discriminated by birds. This assessment is based on the computation of discriminability (ΔS, equations 2 and 3 in Vorobyev & Osorio (1998a)) between the average spectra of supplemented and unsupplemented birds for each different colour patch using D65 as illuminant, the reflectance spectrum of green leaves as background, a Weber fraction ω = 0·05 (Vorobyev et al. 1998b) and relative cone proportions of 1 : 1 : 2 : 2, (VS : S : M : L cones; Endler & Mielke 2005). Discrimination thresholds are set by receptor noise which differs for the cone types and ultimately will determine whether two points in the avian visual space are at least theoretically discriminable or not (Vorobyev & Osorio 1998a; Vorobyev et al. 1998b). ΔS is measured in just noticeable differences (jnds) and values of ΔS > 1 jnd could theoretically be discriminated under these conditions. We re-computed discriminability values between treatments for a series of different backgrounds and the results are identical (data not shown). Similar results have been obtained by Delhey & Peters (2008) and Eaton (2005) for a range of different species.

Previous work on greenfinch coloration has generally used composite scores of colourfulness whereby the size and brightness (visually assessed) of various plumage patches was combined to one or two indexes or principal components (Merilä et al. 1999; Lindström & Lundström 2000). More recently, Saks et al. (2003a), Hõrak et al. (2004) and Karu et al. (2007) assessed colour in male greenfinches from reflectance spectra of tail and breast feathers. They calculated chromavis and huevis, parameters that assess chromatic purity and spectral shape between 400 and 700 nm (excluding the UV), and correlate moderately with canary xanthophyll content of greenfinch feathers (chromavis r 0·3–0·5, huevis r 0·1–0·5 (Saks et al. 2003a)). However, yellow feathers have significant UV reflectance (Fig. S1 in Supplementary Material), and these two parameters therefore capture variation in coloration only incompletely. Nonetheless, in our data set they did correlate with PCxyz(chromavis mean | r | ± SE = 0·57 ± 0·7, huevis mean | r | ± SE = 0·30 ± 0·8) as well as Euclidean chroma (chromavis mean | r | ± SE = 0·71 ± 0·5, huevis mean | r | ± SE = 0·38 ± 0·8), and revealed similar effects of supplementation (data not shown).

blood sampling and analysis

Birds were captured between 9:00 and 17:00 h. Within 2–5 min after capture we punctured the brachial vein, collected two capillaries of blood and stored them vertically on ice. Within 1–2 h we centrifuged the samples for 5 min at 13 000 r.p.m. in a haematocrit centrifuge (Hettich Hematokrit 210) and extracted the plasma (see below). From the first blood sample, the initial sexing was verified by established molecular genetic methods (Griffiths et al. 1998); all birds were confirmed to be male). Plasma samples were kept in the dark and stored at –70 °C until carotenoid analysis by high-performance liquid chromatography (HPLC).

For HPLC analysis and transport to Göteborg, Sweden, 20 µL plasma was removed and mixed with 350 µL acetone (VWR International, Sweden). On arrival, samples were frozen over night at –80 °C and then centrifuged at 260 g for approximately 5 s. The liquid phase was filtered through a 0·2 µm syringe filter (GHP Acrodisc 13 mm, Pall Gelman Sciences Inc., Ann Arbor, MI), and subsequently evaporated in a vacuum centrifuge (Savant DNA 120, Holbrook, USA). The samples were not allowed to dry completely to minimize air exposure and resuspended in 100 µL of the mobile phase (70 : 30 acetonitrile:methanol, v/v, VWR). Depending on the sample colour (i.e. a rough estimate of carotenoid concentration), 20–40 µL of the sample was injected with isocratic mobile phase (see above), through a 100 µL loop into a RP-18 column (ODS-AL, 50 × 4·0 mm i.d., YMC Europe GmbH, Schermbeck, Germany), fitted on a ThermoFinnigan (San Jose, CA) HPLC system with a PS4000 ternary pump, AS3000 auto sampler, and UV6000 diode-array UV/VIS detector. Column temperature was maintained at 30 °C with a flow-rate of 0·6 mL min−1. The run time was set to 15 min. 2D (at 450 nm) and 3D (300–600 nm) chromatograms were obtained and analysed with ChromQuest 4·0 software (ThermoFinnigan). Major pigment fractions were identified and quantified by comparisons to standards of different concentrations of lutein (β,ɛ-carotene-3,3′-diol) and zeaxanthin (β,β-carotene-3,3′-diol), provided by Roche Vitamins Inc., (Basel, Switzerland). Lutein made up 78·7 ± 0·5% of the pool of plasma carotenoids, in addition to 7·1 ± 0·4% Zeaxanthin, 7·0 ± 0·3% LuteinZ, 4·2 ± 0·2% dehydrolutein and ≤ 1% of other carotenoids (Y, Y2 and Zeaxanthin13) each. As lutein was positively correlated with other carotenoid fractions (0·82 < r < 0·92) and with total carotenoid concentration (r = 0·996, all P < 0·0001), we used lutein concentration in the statistical analysis.

statistical analysis

Overall differences in reflectance between plumage patches were compared for pre-moult parameters using anova and Tukey–Kramer post hoc comparisons of all pairs of patches. For each plumage patch we calculated the change in all colour parameters (post-moult minus pre-moult) and examined treatment effects in t-tests. As this analysis identified that tail was different from other traits, we subsequently formally compared sensitivity to the diet treatment in a model with an interaction term trait × diet (analogous to Cotton, Fowler & Pomiankowski (2004a), but with two levels of diet and trait = tail or other). We related mid-moult plasma lutein concentration to post-moult coloration in linear regression analyses. There was no effect of time of day on plasma lutein concentrations (F1,37 = 1·9, P = 0·17, controlling for diet). The effects of general condition during moult on plumage colour were analysed in multiple regression models with mid-moult body mass and haematocrit as explanatory variates, controlling for diet treatment. We did not include tarsus as a covariate as it was not significantly correlated with body mass (including tarsus did not change any results; data not shown). Although six (of 90 pre- and post-moult and change) colour parameters followed distributions slightly but significantly (P < 0·05) different from normal, as this is close to 1 : 20 expected by chance, and as these appeared to be randomly spread between patches and parameters, we used parametric statistics and no transformations. We controlled for multiple testing by implementing the ‘false discovery rate procedure’ (for details see Benjamini & Hochberg 1995; Peters et al. 2007) for each patch for each of these three sets of tests (effects of diet, plasma lutein, general condition). All analyses were performed in jmp in 5·1 (SAS).


The reflectance analysis confirmed human visual assessment that the back and crown plumage is less yellow, that is, less chromatic, than that of the breast and tail, with the rump intermediate. Pre-moult carotenoid chroma (Fig. 1a), chroma and absorptance at intermediate wavelengths (–y) increased from back – crown – rump – tail – breast and although none of the adjacent pairwise comparisons were significant, the back and the crown were always less chromatic than the tail and the breast (see Table S1 in Supplementary Material, for details). The tail showed much higher average reflectance (brightness, 31·4 ± 1·2, P < 0·001) than the other plumage patches (back: 11·7 ± 0·3; crown: 12·4 ± 0·4; rump: 16·2 ± 0·6; breast: 16·4 ± 0·5, see also Supplementary Fig. S1). Reflectance spectra of all plumage patches changed significantly over the moult, mostly irrespective of diet. This change was largely due to a relative increase in UV reflectance (an increase in z was observed in all patches, all t18 > 3·31, all P < 0·004). Additionally, there was a moderate decline in chroma and carotenoid chroma, that was significant for back, breast and crown (all t18 < –2·78, all P < 0·02) but not for rump (both | t18 | < 0·46, P > 0·67) and tail (t18 = −0·10, P = 0·90), the latter due to a difference between treatment groups, as (carotenoid) chroma increased in supplemented males (both t18 > 3·99, P < 0·004). The post-moult plumage of the crown and back was thus less chromatic (lower carotenoid chroma, chroma, higher y) than the tail and the rump, with breast intermediate, and this was irrespective of diet treatment (data not shown).

Figure 1.

Pre-moult colour and effects of lutein supplementation on coloration for the five examined plumage patches of male greenfinches (back, crown, rump, tail and breast). (a) Carotenoid chroma (mean ± SE) before the moult. Letters indicate patches that are not significantly different in Tukey–Kramer post hoc tests (α = 0·05). For differences between patches in other colour parameters, see Supplementary Table S1. (b) Change (mean ± SE) in carotenoid chroma over the moult for males that received a supplementation with lutein during moult (squares) compared to males that received a control diet (diamonds). The effect of supplementation is far greater for the tail than for the other plumage patches (***: P = 0·001, all others ns), for statistical details see Table 1, for effect sizes ± SE see Supplementary Table S2. For post-moult reflectance spectra see Supplementary Fig. S1.

Before the experimental diets began, there were no differences between experimental groups in body mass (t18 = −1·72, P = 0·10), haematocrit (t18 = 1·52, P = 0·14), or any of the reflectance measurements in any of the plumage patches (x, y, z: all | t18 | < 1·27, all P > 0·22); PCxyz: all | t18 | < 0·67, all P > 0·51; chroma: all | t18 | < 1·12, all P > 0·28; carotenoid chroma: all | t18 | < 1·33, all P > 0·20). Birds maintained body mass throughout the experiment (mean change 0·29 ± 0·23 g, t18 = 1·24, P = 0·23) and body mass change did not differ between treatment groups (t18 = 0·42, P = 0·68). Haematocrit declined slightly in the course of the experiment (mean change –0·04 ± 0·01, t18 = −5·75, P < 0·001), presumably due to lower activity levels, but this did not vary between experimental groups (t18 = 0·95, P = 0·36).

Pre-moult plasma lutein concentration did not differ between the two groups (t18 = −0·55, P > 0·6, means 15·3 ± 1·8 and 17·2 ± 2·9 µg mL−1, range 4·4–30·4). In the non-supplemented group, the lutein concentration remained constant (mid-moult 15·4 ± 1·6 µg mL−1, post-moult 13·7 ± 2·2 µg mL−1, F2,27 = 0·26, P = 0·8). Moulting free-living male greenfinches had very similar mid-moult plasma lutein levels (t14 = 0·01, P = 0·99, mean 15·4 ± 2·5 µg mL−1, range 4·6–21·3) to the non-supplemented group, indicating that our non-supplemented diet adequately mimicked the natural diet. On the other hand, there was a significant increase in plasma lutein in the supplemented group (mid-moult 26·4 ± 1·5 µg mL−1, post-moult 32·2 ± 3·6 µg mL−1, F2,27 = 7·31, P = 0·003) resulting in a significant difference in plasma lutein between the two groups mid-moult (t18 = −5·06, P < 0·001) as well as post-moult (t18 = −4·43, P < 0·001). Average lutein levels of the supplemented group were at the peak levels seen in pre-experimental birds (30·4 µg mL−1, see above) and similar to those reported by Tella et al. (2004) for free-living greenfinches in winter in Spain (mean 26·8 µg mL−1, SD = 15·1, n = 8). Therefore it seems that the supplementation resulted in a physiologically relevant increase in lutein availability.

coloration, carotenoids and condition

Supplemented males developed more yellow tails than males fed the standard diet: the treatment significantly affected the change in all colour parameters of the tail (Table 1, see Supplementary Fig. S1, for reflectance spectra and 3D representation of the relative cone quantum catches in the avian visual space). For the breast some colour parameters were affected, notably z, but this was not significant after correcting for multiple testing (Table 1). There was no significant effect of carotenoid supplementation on the change in back, crown and rump coloration for any of the colour parameters (Table 1, Fig. 1b). Averaged across the colour parameters, mean differences between treatments were 2·4–6·6 times greater for the tail than the other plumage patches (see also Supplementary Table S1). Accordingly, sensitivity of the tail to the diet manipulation compared to the other traits (the effect of the interaction between diet and plumage patch on the change in plumage colour) was greater for chroma (F1,96 = 4·99, P = 0·03), PCxyz(F1,96 = 4·17, P = 0·04), y (F1,96 = 3·82, P = 0·05), and z (F1,96 = 3·56, P = 0·06), but it was only marginally greater for carotenoid chroma (F1,96 = 2·15, P = 0·15; power analysis indicated that an additional seven experimental birds would have been needed to make the difference significant) and not significant for x (F1,96 = 0·81, P = 0·37). Finally, average difference between treatment groups in colour of the tail was not only significant but also highly discriminable at 6·14 jnds whereas for other plumage parts, average discriminability between treatment groups was much lower ranging from 3·69 for rump, 3·19 for breast, 1·50 for back to 0·72 for the crown (cf. Supplementary Fig. S1).

Table 1.  Effect of experimental diet (lutein supplementation), plasma lutein concentration and general body condition during the moult on colour parameters for five plumage patches. Significant results are highlighted in bold, those P ≤ 0·05 that are not significant after controlling for multiple testing (FDR, see text for details) are italicized
Effect ofonBackBreastCrownRumpTail
  1. Diet = carotenoid (standard diet supplemented with lutein) / control (standard diet); [lutein] = mid-moult plasma concentration of lutein (µg mL−1); bm = body mass (condition); ht = haematocrit; Δ = change over the moult (post-moult – pre-moult); x, y, z, relative cone quantum catches; PC, PCxyz, first principal component of x, y, z; ECH, Euclidean chroma; CC, carotenoid chroma; t = t18, F = F1,18 for linear regression of [lutein] and F1,16 for multiple regression of bm and ht (controlling for diet). See main text for further details and Supplementary Table S1 for effect sizes and regression slopes.


Mid-moult plasma lutein concentration significantly affected all colour parameters of the post-moult tail (Table 1, Fig. 2a, 0·23 ≤ inline image ≤ 0·29). Breast coloration, notably z, was somewhat positively related to plasma lutein (inline image = 0·23 for z, other inline image≤ 0·11), but this was not significant after correction for multiple testing (Table 1). There was no relationship between mid-moult plasma lutein concentration and colour of the back, rump and crown (Table 1, –0·05 ≥ inline image ≤ 0·04).

Figure 2.

Carotenoid chroma of the newly-moulted contrasting yellow patch in the tail of male greenfinches in relation to mid-moult (a) plasma lutein concentration (b) body mass and (c) haematocrit. Lines are linear regression lines, for (b) and (c) predicted by a multiple regression model also including diet group (for significance details see Table 1, for regression parameters see Supplementary Table S2).

General condition positively affected coloration of the tail (0·47 ≤ inline image ≤ 0·67, GLM including body mass and haematocrit, controlling for diet treatment, see also Table 1): both mid-moult body mass and, to a somewhat lesser extent, haematocrit were positive correlated with the expression of colour (Fig. 2b,c). There were no significant effects of body mass or haematocrit on colour of the rump, back, crown or breast (–0·07 ≤ inline image ≤ 0·32, Table 1).


In this experiment, we supplemented moulting male greenfinches with a moderate amount of carotenoids (lutein), so that the non-supplemented group maintained plasma lutein concentration at the levels found in wild (moulting) birds, while in the supplemented group average plasma lutein gradually increased to around the peak of the natural levels. We compared mean treatment effects as well as the effects of mid-moult circulating lutein and body mass and haematocrit (as indicators of general condition) on reflectance of five carotenoid-based colours, ranging from bright yellow to olive-grey.

Our experiment clearly demonstrated that the contrasting bright-yellow colour of the tail is more responsive to dietary carotenoid content than any other plumage areas. Irrespective of the parameters used to describe the plumage reflectance spectrum, only for the tail did we find that (i) supplemented males developed significantly more colourful plumage (Table 1, Fig. 1b), (ii) colour of the tail correlated with lutein concentration in the plasma during mid-moult (Table 1, Fig. 2a) and (iii) coloration demonstrated significant correlations with body mass and haematocrit (as indicators of general condition, Table 1, Fig. 2b,c). Moreover, direct comparison of sensitivity to the diet showed that tail colour was more responsive to the manipulation than other traits, particularly with respect to the more holistic parameters (Euclidean) chroma and PCxyz, and their component relating to short wavelengths (y and z). Finally, avian visual models confirmed that the average difference due to supplementation was most discriminable for the tail.

While the entire reflectance spectrum of the tail appeared affected by carotenoid availability, for the yellow breast there only was some evidence that relative reflectance in the UV (indicated by z, or relative stimulation of the VS cone) was lowered in the experimental group and declined with mid-moult circulating lutein, although this was not significant after controlling for multiple testing. As carotenoid chroma – an indicator of carotenoid content of the feathers – was not affected, this implies that this could potentially be ascribed to the anti-oxidative actions of carotenoids. For example, reflectance due to deposited melanin could be altered as melanin is a component of olive-green feathers (Lucas & Stettenheim 1972) and as melanin is also sensitive to oxidative stress (McGraw 2005). In two finch species, however, ornamental (McGraw & Hill 2000) as well as non-ornamental (Hill & Brawner 1998) melanin coloration was not sensitive to coccidian infection, whereas this simultaneously affected condition, carotenoid availability and carotenoid-based coloration (for review see Hill 2006).

There was no evidence that carotenoid availability affected the reflectance spectra of the back, crown and rump, although these plumage areas display the trough in reflectance around 450 nm characteristic for xanthophyll absorption (Supplementary Fig. S1). However, the back and the crown and to a lesser extent the rump are less saturated (lower carotenoid chroma, lower chroma, higher y, see also Supplementary Table S1) than the breast and tail. Possibly these less yellow feather tracts contain less carotenoids and possess a more limited ability to locally sequester carotenoids from the blood stream (see McGraw 2006), thereby dampening their responsiveness to carotenoid availability. However, as breast and tail feathers probably contain similar amounts of carotenoids (based on a sample of four greenfinches, Stradi et al. 1995) and are similarly chromatic (Fig. 1a), but respond differently to lutein supplementation, it appears that carotenoid content per se does not (fully) determine responsiveness to carotenoid availability. Yellow-green and olive feathers also contain melanins (Lucas & Stettenheim 1972), and alternatively or additionally, there could be a masking effect of melanin deposition. For example, melanin deposition could co-vary with carotenoid deposition to limit colour variation (see Grether et al. 2005 for an example of such a co-variation process for carotenoids and drosopterins in guppies), a suggestion that could be tested by analysing carotenoid and melanin content of feathers.

Irrespective of the mechanism, colour of the plumage areas other than the bright yellow tail was not significantly sensitive to carotenoid supplementation (Table 1, Fig. 1b), lutein concentration or body mass and haematocrit. Accordingly, discriminability of mean differences between treatment groups was much lower than for the tail (see Supplementary Fig. S1). Nonetheless, at larger sample size possibly treatment differences would have been recorded for these traits, particularly for carotenoid chroma, where all traits showed an increase in response to the supplemented diet (Fig. 2) and the interaction trait × diet was not quite significant. Power analysis however indicated that a much larger sample would have been required to detect significant treatment effects for the other traits, requiring n = 62 (back), 74 (breast), 173 (crown), and 128 (rump) experimental males respectively, highlighting the much higher responsiveness to lutein supplementation of the tail.

The contrasting yellow tail patch, the most sexually dimorphic trait of the greenfinch, is used in sexual displays (Cramp & Perrins 1994) and its colour is an indicator of health and quality (Lindström & Lundström 2000; Hõrak et al. 2004). Additionally, it is the most variable of the greenfinch colours (Delhey & Peters 2008). This combination of characteristics suggests that the tail patch is the most prominent or important of the sexually selected plumage of male greenfinches. Although this suggestions needs confirmation, it is supported by the greater sensitivity of the tail patch to condition and carotenoids compared to the other carotenoid-based plumage patches: a central, but surprisingly ill-supported, assumption of current theories on honest signalling by sexual ornaments is heightened condition-dependence of sexual traits compared to control traits (similar traits under weaker or no sexual selection, Bonduriansky 2007; Cotton et al. 2004b). Our results suggest that the striking yellow tail patch of the male greenfinch is an ornamental trait that appears specifically adapted to honestly signal information on carotenoid availability and general condition of the male during moult.


Authors thank Evi Fricke for help with collecting and processing the spectra, Maria von Post for HPLC laboratory assistance, and Evi Fricke, Monika Krome, Karl-Heinz Siebenrock and Tanja Vogler for help with bird care. We are indebted to Wolfgang Fiedler for logistical support, in particular with the application for permits. Authors are grateful to Andrew Pomiankowski for advice and support, and to Geoff Hill, Peeter Hõrak and the associate editor for insightful comments and suggestions. Authors highly appreciate Peeter Hõrak providing us with access to Eley's PhD thesis. This study was funded by the ‘Sonderprogramm zur Förderung hervorragender Wissenschaftlerinnen’ of the Max Planck Society (to A.P.) and the Swedish Science Council (to S.A.). Author contributions: A.P.: conceived and designed the experiment, analysed data, wrote the paper; K.D.: analysed reflectance data; S.A.: analysed carotenoids; H.N., M.I.F.: planned and performed the research.